zephyr-6b-dpo-full

This model was trained from scratch on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.5676
• Rewards/chosen: -0.5685
• Rewards/rejected: -1.1091
• Rewards/accuracies: 0.7305
• Rewards/margins: 0.5407
• Logps/rejected: -384.2899
• Logps/chosen: -336.4684
• Logits/rejected: -0.9166
• Logits/chosen: -0.9852

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-07
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 2
• total_train_batch_size: 128
• total_eval_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.6548	0.21	100	0.6548	-0.1060	-0.2092	0.6680	0.1032	-294.2936	-290.2218	-0.6577	-0.7018
0.5881	0.42	200	0.5968	-0.3050	-0.6575	0.6875	0.3525	-339.1259	-310.1252	-0.7415	-0.8023
0.5753	0.63	300	0.5734	-0.4917	-0.9768	0.7227	0.4851	-371.0564	-328.7914	-0.7579	-0.8263
0.5602	0.84	400	0.5689	-0.4964	-1.0157	0.7344	0.5193	-374.9490	-329.2605	-0.8579	-0.9263

Framework versions

• Transformers 4.36.2
• Pytorch 2.1.2+cu121
• Datasets 2.14.6
• Tokenizers 0.15.1